How Much Liquidity Insurance Can Lines of Credit Provide? the Impact of Bank Reputation and Lending Relationship*

How Much Liquidity Insurance Can Lines of Credit Provide? The Impact of Bank Reputation and Lending Relationship*

Zhaohui Chen McIntire School of Commerce University of Virginia

Yan Hu University of Minnesota Duluth

Connie Mao Temple University

Current version: February, 2011

JEL Classification: G21, G32

Keywords: Lines of credit, liquidity insurance, bank reputation, lending relationship, moral hazard.

* Zhaohui Zhen, McIntire School of Commerce, University of Virginia, Charlottesville, VA 22903. Tel: (434) 243- 1188; Email: [email protected]. Yan Hu, Department of Finance & MIS, Labovitz School of Business and Economics, University of Minnesota Duluth, Duluth, MN 55812. Tel: (218) 726-7083; Fax: (218) 726-7516; Email: [email protected]. Connie X. Mao, Department of Finance, Fox School of Business and Management, Temple University, Philadelphia, PA 19122. Tel: (215) 204-4895; Fax: (215) 204-1697; Email: [email protected]. We would like to thank Lalitha Naveen, David Reeb, Elyas Elyasian, Warren Bailey, and seminar participants at University of Virginia and University of Minnesota Duluth for their helpful comments and discussions. All errors are solely ours.

How Much Liquidity Insurance Can Lines of Credit Provide? The Impact of Bank Reputation and Lending Relationship

Abstract

Theories suggest that firms use lines of credit as a liquidity insurance to secure a desirable investment level in the event of future downturn (Holmstrom and Tirole,1998; Tirole, 2005). In this paper we examine the liquidity insurance hypothesis by directly quantifying the extent of insurance that a line of credit provides and shed light on how bank reputation and prior lending relationship help a credit line to supply more efficient liquidity. Consistent with the liquidity insurance hypothesis, we find that firms are more likely to use credit lines at times of poor performance, and the drawdown rate is on average significantly lower than the imputed market cost of borrowing given the firm's financial condition at the time of drawdown. In addition, we document that the strength of prior lending relationship is associated with a lower drawdown rate. Furthermore, the impact of prior lending relationship on the drawdown rate only exists for borrowers subject to greater information asymmetry. While borrowers are penalized (paying a higher loan spread and annual fee and more likely to pledge collateral) on new lines of credit issued after their drawdown, the penalty is smaller as they borrow from higher reputation banks. Our results suggest that bank reputation and prior lending relation both help provide a more efficient liquidity insurance, however through different channels.

1. Introduction

A line of credit is a bank's promise of future lending sold to a borrower.1 The borrower may draw down at any time prior to maturity, up to the maximum amount, paying interest at a pre-determined spread over LIBOR or prime rate specified by the agreement. Lines of credits are the most popular form of bank lending, representing 80% of commercial loans in the United

States (Duca and Vanhoose 1990).2 According to FDIC (www2.fdic.gov/SDI), outstanding unused lines of credit of U.S. corporation amount to $1.7 trillion at the end of 2004.

The literature offers two hypotheses for why firms use line of credit. One is the convenience hypothesis, under which a credit line assures a firm a certain level of funding in the future. This convenience is valuable if the firm faces uncertainty about its funding needs. An example is when the firm is exploring acquisition opportunities in a timely fashion.3 The other hypothesis is the liquidity insurance hypothesis. A line of credit can provide funding so that a firm can take positive NPV project when the firm cannot otherwise get financing due to frictions in the financial markets. For example, Holmstrom and Tirole (1998) and Tirole (2005, 2010) argue that in the absence of a line of credit, a firm may have to forgo positive NPV projects in a future downturn. This is because after the firm gives the minimum cash flow from the project to the entrepreneur to motivate him to work hard, there is not enough left-over cash to be pledged to

1 Borrowers usually pay a upfront fee as well as an annual commitment fee for the option to access liquidity. 2 In the Dealscan data, 63% (or 73%) of loans are lines of credit based on the number of loans (or the amount of loans). These number increases to 82% (or 86%) in our sample that are present in Compustat and CRSP, and tend to be larger firms. 3 Lins, Servaes, and Tufano (2010) analyze survey data collected from CFOs of public and private firms in 29 countries and document that 60% of firms view that lines of credit provide certainty of funding during event risk or acquisition opportunities, and 32% of firms indicate that the time to raise funds is an important consideration as they use lines of credit. 1 the investors to raise the necessary funding for the project.4 Note that these two hypotheses are not mutually exclusive.

The existing empirical evidence suggests a limit to the liquidity insurance hypothesis.

Campello, Giambona, Graham, and Campbell (2010) find that firms that are more likely to need lines of credit (small, private, noninvestment grade, or unprofitable) have less access to credit lines than their large, public, investment-grade, profitable counterparts. Sufi (2009) finds that lines of credit are reduced when a firm‟s liquidity level is low. If the purpose of a line of credit is to insure a firm against liquidity shock, why is it withdrawn when the firm most needs it? A bank has at least two ways to renege on a line of credit commitment: renegotiate regarding credit line terms such as loan rate, or revoke the entire loan commitment if the firm‟s financial condition worsens by invoking the Material Adverse Change (MAC) clause.5,6 Given the lender's many options to escape the commitment, to what extent, if any, does a line of credit provide liquidity insurance?

The first goal of our paper is to address this question. First, we examine the difference between the loan rate at which a firm draws down a line of credit and the rate the firm could obtain in the lending markets at the time of drawdown. The extent of liquidity insurance would be measured by the difference between the drawdown rate and the market lending rate. We find evidence supporting the liquidity insurance hypothesis: on average the drawdown rate is about

25 basis points lower than the imputed market rate given the firm‟s financial condition. For firms with a strong prior lending relationship with the bank, the drawdown rate is 63 basis points lower

4 See also Boot, Thakor, and Udell (1987, 1991) and Berkovitch and Greenbaum (1991) for other frictions that a line of credit can mitigate to reduce under-investment problem. 5 Roberts and Sufi (2009) show that over 90% of long-term debt contracts are renegotiated prior to their stated maturity. Renegotiation leads to significant changes the maturity, amount, and spread of the contract. Less than 18% of renegotiations are directly or indirectly linked to a covenant violation or payment default. 6 Most loan commitment contracts include a MAC clause, which permits the bank to decline to lend under the commitment if the borrower‟s financial condition has declined significantly since the commitment was sold. 2 than the imputed market rate. On the other hand, we find that the drawdown rate is higher than the contract rate, indicating renegotiation between the firms and banks and/or performance pricing determine the drawdown rate. Furthermore, more than 50% of the lines of credit in our sample are drawn down, suggesting that line of credit provides economically significant liquidity insurance to the firm. Finally, consistent with the insurance hypothesis, a firm is more likely to draw down when it experiences a negative shock as measured by ROA or violations of debt covenants.

If a line of credit provides insurance to a firm experiencing a financial downturn, it is subject to the classic moral hazard problem extensively studied in the insurance literature (for example, Rothschild and Stiglitz, 1976). More specifically, the firm may be less careful in avoiding a financial downturn because a credit line guarantees access to relatively cheap financing.7 For lines of credit to add value, they must feature mechanisms that motivate borrowers to behave appropriately. One such mechanism in the literature is future punishment

(Radner, Myerson, and Maskin, 1986; Atkeson and Lucas, 1995). More claims filed by the insured would lead to a higher premium in the future. We study this type of mechanism by investigating whether borrowers tend to be punished in the future after drawing down a line of credit. We find that a drawdown leads to higher rates and greater collateral on future lines of credit. These results hold after we control for the firm‟s characteristics at the time when the new line of credit is granted. This result supports the liquidity insurance story. An alternative form of punishment, denial of any future new line of credit, is inefficient in that it not only hurts the firm but also the bank because the bank loses business. We find that banks seldom use this form of punishment of drawdown firms.

7 Alternatively the firm may undertake negative-NPV projects to take advantage of the cheap financing as shown by Holmstrom and Tirole (1998). 3

Both borrowers and lenders are subject to moral hazard problems in their management of lines of credit. As mentioned previously, the bank can renege on its commitment and the firm can take higher risk to exploit the liquidity insurance the bank provides. As a result, the insurance the line of credit provides is incomplete. First, a drawdown at a lower-than-market rate may lead to a higher rate on future lines of credit. Second, the benefit a firm can obtain from drawdown cannot be too large or the bank may renege on it.8 Therefore, a line of credit provides at best partial insurance. The second goal of this paper is to study factors that may mitigate these moral hazard problems. In particular, we investigate how the lender‟s reputation and the prior lending relationship between the lender and the borrower contribute to the efficiency of the liquidity insurance.

We find that both bank reputation and prior lending relationship help improve the efficiency of liquidity insurance. A strong bank-borrower relationship lowers the drawdown rate without increasing post-drawdown punishment. A high bank reputation (measured by the bank‟s market share in lending) lowers post-drawdown punishment without increasing the drawdown rate. In other words, both factors enhance the liquidity insurance contract by increasing the amount of insurance for the same amount of punishment.

Our paper contributes to the literature on lines of credit on several dimensions. First, to the best of our knowledge, this is the first paper that studies the actual drawdown rates on lines of credit. As a result, we can directly measure the extent of the insurance that a line of credit provides as suggested by the theory.9 Secondly, our study adds to the empirical literature by

8 The effect largely disappears if the players are "patient" (the folk theorem of game theory, as in Fudenberg, Levine, and Maskin, 1994). However, if players are not patient, inefficient outcomes can arise (see Levin (2003) for an analysis in a principal-agent framework). 9 See Boot, Thakor, and Udell (1987), Berkovitch and Greenbaum (1991), Holmstrom and Tirole (1998), and Tirole (2005, 2010). 4 providing an analysis of the firms‟ drawdown events.10 Finally, we shed light on how bank reputation and prior lending relationship can help a line of credit contract to provide better liquidity insurance.

The rest of this paper proceeds as follows. Section 2 reviews the literature and develops our hypotheses. Section 3 discusses data and methods. Section 4 presents empirical results and section 5 concludes the paper.

2. Literature Review and Hypotheses Development

2.1. Lines of Credit and Liquidity Insurance

The theoretical literature argues that lines of credit are committed liquidity insurance that can overcome capital market frictions. Boot, Thakor, and Udell (1987) argue that a line of credit may be thought of as a put option for the firm to borrow at the pre-arranged low rate if the spot- market interest rates are high. Berkovitch and Greenbaum (1991) show that a line of credit contract may resolve the underinvestment problem of Myers' (1977). Holmstrom and Tirole

(1998) and Tirole (2005) argue that a line of credit contract mitigates the under-investment problem that arises in that an entrepreneur will not exert sufficient effort without a minimum share in a project.

Despite the well-established theoretical notion that lines of credit serve as fully committed liquidity insurance, empirical support is rather limited. Sufi (2009) finds that firms with low cash flow are less likely to obtain a line of credit and credit lines are revoked when

10 Sufi (2009) and Campello, Giambona, Graham, and Campbell (2010) focus on how firm characteristics are related to the amount of lines of credit available to the borrowers and the amount of their drawdown. Jimenez, Lopez and Saurina (2009) examine the determinants of the usage of lines of credit with a Spanish dataset. They find that the use of a line of credit increases with the default probability of the line of credit, however decrease with the firm's prior default history. Longer bank relationship leads to lower extent of drawing down lines of credit. Nevertheless they do not have drawdown rate information and cannot directly quantify the liquidity insurance benefit, and they did not analyze post-drawdown punishment. 5 firms‟ liquidity levels are low. Campello, Giambona, Graham, and Campbell (2010) use survey data during the 2008-2009 financial crisis, and find that small, private, low credit rating or unprofitable firms draw more heavily on their lines of credit, at the same time they are more likely to face difficulties renewing or initiating lines of credit. Jimenez, Lopez and Saurina (2009) examine the determinants of the usage of lines of credit by Spanish firms. They find that drawdown increases with the probability of default, but decreases with the firm's prior default history and the length of the relationship with its lenders.

As discussed above, the theoretical literature argues that a line of credit can guarantee relatively cheap financing if a firm faces future liquidity shock. This motivates the following hypothesis:

H1: The drawdown rate is lower than the market rate the borrower is able to obtain in the lending market at the time of drawdown.

Additionally, both borrower and lender may have incentives and means to exploit the line of credit contract, so its liquidity insurance the contract provides may not be efficient. In particular, to prevent the borrower from taking excess risk and potentially incurring unnecessary negative shocks, the bank must punish firms that draw down a credit line. Therefore, we have the following hypothesis:

H2: All else equal, a borrower is penalized with a higher rate and more stringent contract terms on new lines of credit after a drawdown.

2.2. Lending Relation, Bank Reputation, and Lines of Credit

It has been argued in the literature that the long-term borrower-lender relationship and lender reputation could mitigate the moral hazard problem of the borrowers as well as the banks.

Prior studies document that lending relationships increase credit availability (Petersen and Rajan,

1994), decrease loan rate and reliance on collateral (Berger and Udell, 1995), and reduce the costs of financial distress (Hoshi, Kashyap, and Scharfstein, 1990). In our context, a strong prior lending relationship allows the bank to gather more and better information about the borrower over time at a lower cost. This enables the bank to monitor the borrower more effectively, thus reducing the moral hazard problem of the borrower. For example, if the bank can effectively prevent a borrower from taking excessive risk, the bank need not punish the borrower for drawing down a line of credit. This is because the bank knows that the borrower draws down because of a liquidity shock rather than to game the contract.

While bank monitoring is essential to mitigate the moral hazard of borrowers, it also gives banks discretion in whether to uphold the contract, at what rate to honor the lines of credit, and how to punish a borrower after drawdown. This inevitably subjects the bank to moral hazard which, however, may be mitigated if the bank wishes to maintain its reputation (Boot,

Greenbaum, and Thakor, 1993; Fudenberg and Levine, 1992; Chemmanur and Fulghieri, 1994).

As a result, to preserve or enhance reputation as an effective monitor, more reputable banks have a greater incentive to monitor borrowers even if it is costly. Similarly, more reputable banks have a greater incentive to grant the drawdown benefit to borrowers or impose less severe post- drawdown punishment to preserve or enhance reputation of "fairness".11 We thus propose the following two hypotheses:

11 A good reputation allows the bank to charge higher prices or gain market share. 7

H3: All else equal, lenders will provide better liquidity insurance ex post to the borrower with which they have a stronger prior lending relationship.

H4: All else equal, more reputable banks will provide better liquidity insurance ex post.

Since firms with greater information asymmetry suffers more from moral hazard, the effect of prior lending relationship and bank reputation should be larger, which leads to our next hypothesis:

H5: All else equal, the effect of lending relationship and bank reputation on the liquidity insurance ex post is greater for borrowers that are subject to greater information asymmetry.

3. Data and Method

3.1. Data and Sample Construction

We first hand match all the borrower names in the Dealscan database during 1990-2007 with Compustat firm names. The Dealscan database contains detailed information on bank loans

(including term loans and lines of credit) worldwide, such as borrower and lender identity, loan amount, spread over LIBOR, issuing and maturity date, financial and general covenants, etc.12

Among the list of matched firms between Dealscan and Compustat, we randomly selected 800 firms. We focus on 800 firms because of the time of gathering and recording information about drawing down lines of credit which is described below.

Through a variety of regulations, the SEC requires firms to file details of material debt agreements, sources of liquidity, and long-term debt schedules (Kaplan and Zingales, 1997; Sufi,

2007; Nini, Smith and Sufi, 2009). For each of the 800 randomly selected firms, we look up 10-

12 About 60% of Dealscan data are collected from SEC filings, and the rest are obtained from direct contact with borrowers and lenders. According to Carey and Hrycay (1999), the Dealscan database covers between 50% and 75% commercial loans in the U.S. by 1992, and by 1995 it covers a greater fraction. 8

K filings at each fiscal year end during 1996-2005. Since we cannot find any 10-K filing for seven firms, we are left with 793 random firms.13 We end up with a panel of 6859 firm year observations. From the 10-K filings, we gather information on whether there exist lines of credit, whether the credit facilities were drawdown and at what rate the credit facilities were drawn down. Information on whether the firm was out of compliance with any financial or general covenants was obtained from the web link http://faculty.chicagobooth.edu/amir.sufi/ (Nini,

Smith, and Sufi, 2009). Since the SEC does not require firms to report the exact drawdown rate of a line of credit, 88.26% of the firm year observations do not state the exact drawdown rate though they indicate one or more lines of credit were drawn down.

To detail the credit facilities being drawn down, we try to match each of the lines of credit in 10-Ks with a contract recorded in the Dealscan database, based on starting date, maturity date, size of the facility, and lender information. However some firms did not offer enough information in the10-K filings to match to Dealscan. As a result, our final sample consists of 804 observations with both drawdown rate available from 10-Ks and detailed contract information available from the Dealscan.14 All facility specific variables are obtained from the

Dealscan database. Other firm characteristics variables are constructed using the

Compustat/CRSP merged database.

3.2. Predicted Drawdown Rate

Based on Tirole (2005), lines of credit would provide insurance to firms and, thus, the drawdown rate should be lower than “the fair market rate” the firm would otherwise pay. To test

13 While most of drawdown rate were expressed in spread over LIBOR rate, some were reported as a percentage of return. For these cases we use the corresponding LIBOR rates to back out the drawdown rate spread over LIBOR. 14 Some observations represent multiple drawdown of the same facility. The drawdown sample consists of 456 facilities issued by 257 firms. 9 the "insurance" hypothesis, we compare the actual drawdown rate with the “the fair market rate.”

Because “the fair market rate” is unobservable, we follow Graham, Li and Qiu (2008) and use the following empirical model to impute a bank loan rate given firm characteristics at the time of drawdown:

Loan spread

 Drawdown dummy, Covenant violation dummy, Firm characteristics, (1) = f  .  Loan characteristics, Industry effects, Macroeconomic factors 

In the regression, each observation represents a single loan. The dependent variable is the loan spread, which is the Dealscan data item all-in drawn spread. We follow Graham, Li and Qiu

(2008) and control for firm characteristics (Log(assets), market-to-book ratio, Leverage, ROA, tangibility, cash flow volatility, and Zscore), loan characteristics (Log(loan size), Log(loan maturity), performance pricing dummy, security dummy, loan type dummies, and loan purpose dummies), industry effects (dummies for each two digit SIC code), macroeconomic factors

(credit spread and term spread), and dummies for each calendar year. In addition, we include in the above equation two dummy variables for drawdown and covenant violation to capture the private information associated with a firm's decision to draw down a line of credit and whether a firm is in compliance with debt covenants. Drawdown is a dummy variable that equals one if a firm draws down one or more credit facility in a particular year, and zero otherwise. Covenants violation is a dummy variable that equals one if a firm violates its debt covenants in a particular year, and zero otherwise. All other variables are defined in Appendix A.

We first estimate equation (1) using all the bank loans issued by the 800 sample firms during 1996 and 2006, and obtain coefficient estimates of all the variables.15 For each drawdown event, the “imputed” loan spread is the regression fitted value based on the coefficient estimates and the firm and loan characteristics at the time of drawdown. This "imputed" rate could be thought as the "fair market rate" a firm could otherwise obtain in the lending market, given its financial condition at the drawdown. We call this "imputed" loan spread as "predicted market rate." The difference between the drawdown rate and the predicted rate is considered as the drawdown rate benefit.

3.3. Measures of Bank Reputation and Lending Relation

Our analysis on bank reputation and prior lending relation focuses on lead arrangers.16

Following Sufi (2007), we classify lenders listed in “Lender - Lead Arranger” as lead arranger if this variable is available on Dealscan. Otherwise, we classify lenders having a “lead role” listed in “Lenders - All lenders” as the lead arranger. To make data collection manageable, we focus on the top 100 lead arrangers listed by Sufi (2007). This selection will not result in any bias because, according to Sufi (2007), the top 100 lenders represent about 96% of the total number of loans.

To account for bank mergers during our sample period, we track all mergers and acquisitions of financial institutions, and allow the acquiring banks to inherit the lending history of an acquired bank after the acquisition date. For example, in April 1998, First Union Corp. acquired

CoreStates Financial Corp. with a name of merged entity „First Union Corp.‟ Thus after April

15 Ideally we would like to estimate the above loan pricing model use all the bank loans of all the firms recorded in Dealscan, however we do not have information on whether a firm is drawing down its credit facility except the 800 randomly select firms. 16 While most loans in Dealscan involve several lenders, the lead arranger is responsible for negotiating directly with the borrower for all contract terms. Other participant lenders rarely directly negotiate with the borrower and usually hold a relatively small share of the loan. 11

1998, First Union Corp. inherited CoreStates‟ entire lending history as we compute lender market share and lead bank-borrower relationship.

Following previous literature (for example, Bharath, Dahiya, Saunders, and Srinivasan,

2007), we use a lead arranger‟s market share in lending in the previous five years as a proxy for lead arranger reputation. It is computed as the dollar amount of loans arranged by a particular lead bank during the previous 5 years divided by the total amount of loans issued in the market in the same period.17 Prior lending relationship is computed as the fraction of all loans received by a particular borrower during the previous 5 years that were arranged by a particular bank.

Specifically, for borrower j in year t, the amount of loans arranged by bank i (and its predecessors) during the previous 5 years is divided by the total amount of loans taken by borrower j during that period. We also construct an alternative measure of the existence of a prior lending relationship, a dummy variable equal to one if a particular bank has ever lent as a lead arranger to a particular borrower during the previous 5 years. By H3 and H4, we expect that prior lending relationship and bank reputation are positively related to the likelihood of drawdown, and negatively related to the drawdown rate.

3.4. Regression Model

We are interested in two related questions. First, how do prior lending relationship and bank reputation affect the likelihood of granting a drawdown of the line of credit upon borrower' liquidity need? Second, conditional on the drawdown, how do lending relationship and bank reputation affect the drawdown rate benefit? Since the choice of drawing down is likely to be

17 We also constructed lead arranger‟s market share in previous year as a proxy for lender reputation, and obtained similar results. 12 correlated with the rate benefit, we adopt the Heckman's (1979) two-stage procedure to model these two questions as follows.

Drawdown dummy

 Bank reputation (or Lending relation), Covenant violation dummy,   (2A) = f  Industry adj - ROA,Cash holding, Large firm dummy, ,    Investment grade dummy, Market-to-book 

Drawdown rate  Predicted market rate  Bank reputation (or Lending relation), Covenant violation dummy,    First Drawdown dummy, Log(asset), Market-to-book, Leverage,  (2B) = f  Industry adj - ROA, Tangibility, Zscore, Log(loan maturity), .    Log(loan size), Performance pricing dummy, Loan type dummies,     Loan purpose dummies, Industry dummies, Inverse Mills Ratio s 

Equation (2A) explains the choice of whether or not to draw down, following Campello,

Giambona, Graham, and Campbell (2010) and including firm-specific variables like cash holding, a large firm dummy for firms with net sales of at least $1 billion, an investment grade dummy, and market-to-book ratio for investment opportunities. We also include a covenant violation dummy and industry-adjusted change in ROA from the previous year to proxy for negative shocks in firm performance. Please see Appendix A for details on variable definitions. The insurance hypothesis suggests that poorly performing firms are more likely to draw down their lines of credit. In addition to variables intended to capture firm demand for drawing down lines of credit, we include variables that proxy for bank reputation and prior lending relationship to model the supply of liquidity. Equation (2B) follows Graham, Li, and Qiu (2008) and includes many firm and loan characteristics to explain the drawdown rate benefit, that is, the difference between the drawdown rate and the predicted market rate. Bank reputation and prior lending

13 relationship are also included. In addition, the inverse Mills ratio from equation (2A) is included to correct for any potential selection bias resulting from the drawdown decision.

4. Empirical Results

4.1. Sample Statistics

Table 1 presents summary statistics. We start with 800 randomly-selected companies during 1996 to 2005. About 83% of total firm-year observations have a line of credit, and 53% of them are associated with a drawdown.18 The mean value of the covenant violation dummy indicates that about 7% of the observations are associated with a covenant violation. Average firm assets is $1.5 billion, while its median is smaller, $277 million. The average leverage ratio of our sample firms is 0.24, average market-to-book ratio is 1.9, and average ROA is 0.097.

4.2. Univariate Analysis

Table 2 presents univariate analysis of a sample of credit facilities for which we are able to identify the drawdown rate from the 10K filings. As shown in Panel A, we are able to identify

804 drawdown rates related to 456 facilities being drawn down by 257 companies. The mean

(median) drawdown rate on lines of credit reported in 10Ks is 196 (175) basis points over

LIBOR. The mean (median) contract rate on lines of credit that were subsequently drawn down is 175 (150) basis points over LIBOR, which is lower than the mean (median) of the actual drawdown rates. The predicted market rate imputed from the regression model has mean (median) of 223 (222) basis points over LIBOR, which appears higher than the mean (median) drawdown rate. We also identify the first new line of credit initiated after each drawdown, and their mean

18 Out of those 800 companies, 750 companies (94.58%) have obtained at least one line of credit during the sample period. 647 firms (86.27%) drew down lines of credit at least once, and the other 103 companies never drew down any line of credit in the entire sample period. 14

(median) contract rate is 177 (150) basis points over LIBOR.19 The average upfront fee and annual fee on the line of credit being drawn down are 41 basis points and 22 basis points respectively. These fees are similar for credit lines issued after the drawdown.20

In Panel B, we compare the actual drawdown rate and the predicted market rate. The drawdown rate is on average 25 basis points lower than the predicted market rate, and this difference is statistically significant at the 1% level. The median difference between the two is 41 basis points, which is highly significant. This evidence supports our hypothesis H1, that lines of credit provide a lower rate when borrowers have liquidity needs. As we compare the drawdown rate with the contract rate on the line of credit being drawn down, the mean difference is 21 basis points and highly significant, though the median difference is not statistically significant. Thus, we find some evidence that actual drawdown rates are higher than the initial contract rate specified at the time when the lines of credit are initiated. In addition, we compare the drawdown rate and the contract rate on new lines of credit established after a drawdown, and find that the mean and median difference between the two are both positive but insignificant.

4.3. Bank Reputation, Lending Relationship, and Drawdown Rate

As argued by hypotheses H3 and H4, due to monitoring advantages and reputation concerns, stronger prior lending relationship and higher bank reputation should be associated with better liquidity insurance ex post, that is, a larger drawdown rate benefit. To test these hypotheses, we first use a univariate test of how the drawdown rate benefit (the difference between drawdown rate and predicted market rate) is related to bank reputation and prior lending relationship. In

Table 3, we sort the sample into two groups based on median bank reputation, median strength of

19 Since some borrowers have not obtained any new line of credit after the drawdown, we have only 631 observations on contract rate of new line of credit. 20 Please note that there are less than 20% of facility contracts containing information on these fees. 15 prior lending relationship, or the existence of a prior lending relationship. After sorting by bank reputation, the mean and median difference between drawdown rate and the predicted market rate are negative and significant in both groups. Moreover, the drawdown rate benefit is not significantly different between the high and low reputation groups. This suggests that bank reputation does not affect the drawdown rate benefit.

Partitioning by prior lending relationship, we find that the difference between drawdown rate and predicted market rate is significantly negative, but only for the subsample with strong prior lending relationship. For this group, the drawdown rate is on average 62.6 basis points lower than the predicted market rate. In contrast, for firms borrowing from banks with which they have a weak prior lending relationship, the drawdown rate is not significantly different from the predicted market rate. The difference between the strong and weak lending relationship groups is negative and highly significant. We obtain similar results when we divide the sample based on the “existence of prior lending relationship” dummy. Again, the drawdown rate is significantly lower than the predicted market rate only for the subsample with prior lending relationship. These results support hypothesis H3 which predicts that lenders offer a lower drawdown rate for borrowers with which they have a stronger prior lending relationship.

Since the choice of drawdown as well as the drawdown rate might be jointly determined by firm and contract characteristics, we next examine H3 and H4 in a multiple regression framework. We estimate equation (2A) and (2B) using the Heckman (1979) two-stage procedure for a sample of firm-years with lines of credit available. That is, observations with and without a drawdown are both included. However, to be included in the estimation, an observation without a drawdown must have non-missing values for all the independent variables in equation (2A). An observation with a drawdown must have non-missing observations for drawdown rate, predicted market rate, and all of the independent variables in both equation (2A) and (2B). As a result, we 16 are left with about 1,200 firm-year observations. Estimation results are reported in Panels A and

B for equation (2A) and (2B) respectively in Table 4.

Panel A examines how bank reputation and prior lending relationship affect the likelihood of drawing down lines of credit. Since we focus on the supply side (that is, the likelihood that banks honor credit lines), we must control for the demand side, (that is, how much borrowers would like to draw down). We find that firms with covenant violations (that is, poor performance) are more likely to draw down lines of credit. Furthermore, the industry-adjusted change in ROA is significantly negatively related to the probability of drawdown. These results are consistent with the insurance hypothesis. In addition, cash holding is significantly negatively related to drawdown, suggesting a substitution effect between cash holding (internal liquidity) and drawing down lines of credit (external liquidity).21 As with Campello, Giambona, Graham, and Campbell

(2010), we find that large firms are less likely to draw down, since they in general have better liquidity than small firms. Neither firm rating (investment grade dummy) nor market-to-book ratio is significantly related to the likelihood of drawdown.

After controlling for the demand for drawdown, neither high bank reputation nor strong lending relationship is significantly related to the probability of drawdown. The coefficient estimate on the existence of prior lending relationship reported in model (3) is negative, though only marginally significant at the 10% level. Model (4) includes dummy variables for both bank reputation and strong lending relationship, and yields similar results. These findings suggest that bank reputation and prior lending relationship do not affect the probability of banks honoring their lines of credit.

21 Campello, Giambona, Graham, and Campbell (2010) also document a significant negative relationship between cash holding (as well as cash flow) and the amount of drawdown during the 2008-2009 financial crisis. 17

The match between banks and borrowers is not random, but might be determined by certain characteristics of borrowers. For example, the theory of Dinc (2000) shows that higher reputation banks offer loans to the highest quality borrowers, though an increase in credit market competition enhances banks‟ incentive to lend to borrowers in distress. As a result, the effect of bank reputation and prior lending relationship on the likelihood of drawdown or drawdown rate

(as we will discuss next) can be driven by borrower characteristics that affect the demand side of drawdown or drawdown rate. To address the endogenous choice between borrowers and banks, we employ a two step-instrumental variable approach following Bharath, Dahiya, Saunders, and

Srinivasan (2007), and include firm size, ROA, asset tangibility, and a dummy for accessing public debt market as instruments.22 Results corrected for the endogeneity issue are reported for models (5) and (6). As with we find above, neither bank reputation nor lending relation is significantly related to the probability of drawdown.

In Panel B, we measure, conditional on the drawdown, how bank reputation and prior lending relationship affect the drawdown rate benefit. The dependent variable in Panel B is the drawdown rate minus the predicted market rate. The more negative the dependent variable, the greater the rate benefit. After controlling for firm and contract characteristics, model (1) shows that high bank reputation dummy is positively related to the drawdown rate, though the result is marginally significant at the 10% level. As shown in models (2) and (3), the strong lending relationship dummy and the dummy for the existence of prior lending relation are associated with a significantly greater rate benefit of 51 and 52 basis points, respectively. Put another way,

“relationship banks” allow their borrowers to drawdown lines of credit at 52 basis points lower than non-relationship banks. When we include both bank reputation and lending relation

22 We define that a firm has access to the public debt market if any type of S&P debt rating is available on Compustat. 18 variables in model (4), the high bank reputation dummy is positive and insignificant, while the strong lending relationship dummy is still negative and highly significant. In models (5) and (6), we obtain similar results after accounting for endogeneity given the choice between banks and borrowers.

In summary, we find support for hypothesis H3: strong prior lending relationship between banks and borrowers results in a significantly larger drawdown rate benefit. Therefore, banks appear to provide better liquidity insurance ex post, in terms of rate benefit, to borrowers with which they have a strong prior lending relation. However, we do not find evidence that more reputable banks offer a greater drawdown rate benefit. As we discuss above, better (or more efficient) liquidity insurance manifests with not just a drawdown rate benefit, but also the extent of post-drawdown punishment. We will examine the latter in Section 4.4.

Among the control variables, covenant violation is associated with a significantly higher drawdown rate of 40-60 basis points. The coefficient estimate on the “first drawdown” dummy is insignificant, suggesting no difference between first and subsequent drawdowns. Borrowers with higher leverage pay a higher rate. Other firm characteristics are insignificant. In addition, larger loan size and longer time to maturity of the line of credit contracts are associated with significantly higher drawdown rates. The coefficient estimates on the inverse mills ratio are insignificant, suggesting that sample election bias is not a major concern for our analysis.

As we discussed in Section 2, prior lending relationship and bank reputation could enhance monitoring and hence mitigate information asymmetry and moral hazard in these contracts. Thus, the effect of lending relationship and bank reputation should be larger for borrowers subject to greater uncertainty and information asymmetry, as suggested by hypothesis

H5. If there is no information asymmetry, the moral hazard problems of the borrower and the

19 bank in gaming the liquidity insurance contract are minimal, thus lending relationship and bank reputation play little role in how likely and how well a contract is honored.

To examine the hypothesis H5, we introduce two measures of uncertainty and information asymmetry, and their interaction terms with bank variables. The first, following Sufi (2009), is the no public debt dummy, equal to one if a firm does not have any type of S&P debt rating on

Compustat. The second one follows Kroszner and Strahan (2001), is a high return volatility dummy that is equal to one for firms with stock return volatility above the sample median. High return volatility suggests high uncertainty and greater information asymmetry, for which the advantages in monitoring and information production due to reputational concerns and relationship lending should be greater. The interaction term, Strong lending relation*No public debt, should capture the differential effect of lending relation between borrowers with high and low information asymmetry. By hypothesis H5, the coefficient estimate on the interaction term should be significantly negative.

Table 5 presents the Heckman two-stage model including interaction terms. Panel A models the decision to draw down or not. As with Panel A of Table 4, the coefficient estimate on high bank reputation dummy is insignificant in model (1) and (2). As expected, the No public debt dummy itself is significantly positively related to the likelihood of drawdown. This is because firms without access to public debt market are more likely to be financially constrained, which then leads to more frequent use of their lines of credit. However, high return volatility dummy is not significantly related to the probability of drawdown. Neither of the interaction terms, High bank reputation*No public debt or High bank reputation*High return volatility is significant. Models (3) and (4) show that strong lending relationship is not significantly related to the likelihood of drawdown, regardless of whether borrowers face more or less severe information asymmetry. In summary, bank reputation and prior lending relation do not affect 20 banks' probability to honor their loan commitments, even for borrowers face greater information asymmetry.

Panel B explores the potential differential impact of bank characteristics on drawdown rate benefit across firms with greater or lesser information asymmetry. As shown by models (1) and (2), the high bank reputation dummy is not significantly related to the rate benefit. While the dummy for no public debt is insignificant, high return volatility is significantly and positively related to the difference between the drawdown rate and the predicted market rate, suggesting a smaller rate benefit. Again, neither of the interaction terms, High bank reputation*No public debt or High bank reputation*High return volatility is significant. Thus bank reputation does not affect the rate benefit conditional on the drawdown, regardless of whether the firm has high or low information asymmetry. In examining the impact of lending relationship in models (3) and

(4), we find that the coefficient estimate on strong lending relationship dummy becomes insignificant as we add two additional variables, the No public debt dummy (or High return volatility dummy) and their interaction terms with strong lending relation dummy. Most interestingly, both interaction terms, Strong lending relation*No public debt and Strong lending relation *High return volatility are negative and significant in models (3) and (4), respectively.

This indicates that, for borrowers with low level of information asymmetry, strong prior lending relationship has no significant effect on the insurance benefit (a lower drawdown rate than the fair market rate). In contrast, for borrower with great information asymmetry, strong prior lending relationship is associated with a significantly lower difference in the drawdown rate and the predicted market rate (that is, a higher rate benefit upon the drawdown). This result is consistent with hypothesis H5: the impact of prior lending relationship is greater for borrowers with more severe information asymmetry, since relationship lending mitigates information problems. 21

4.4. Penalty after the Drawdown of Lines of Credit

As discussed above, moral hazard arises because a firm might be less careful in managing a financial downturn or take excessive risk since a line of credit guarantees access to cheap financing. To mitigate opportunistic drawdown behavior, banks would impose penalties after drawdown when a borrower tries to renew or obtain new lines of credit.

4.4.1. Effect of Drawdown on the Likelihood of Obtaining a New Line of Credit and the Rate

Change

Banks can penalize borrowers in two ways after a drawdown: the line of credit may not be renewed, or a higher rate can be charged on a new credit line. To investigate whether and how banks penalize borrowers via these channels, we examine the effect of drawdown on the likelihood of obtaining a new line of credit as well as the change in loan contract rate conditional on obtaining a new credit line. For this purpose, we start with all the lines of credit in the

Dealscan database issued by our 800 randomly selected firms. For each line of credit, we try to identify whether the firm issued a new line of credit within two years after the expiration of the existing credit line. If so, the "Newline" dummy is set to one, zero otherwise. The

"Newline_increase" dummy is equal to one if the contract rate on the new line of credit is higher than that of the existing one, zero otherwise. We examine the post-drawdown penalty with a

Heckman probit model framework. In equation (3A), the dependent variable in the selection model is the dummy Newline. Conditional on having a new line of credit, we examine how the drawdown event affects the likelihood of a rate increase with equation (3B).

Newline dummy  β  β Drawdown dummy  β Covenant violation dummy t 0 1 t1 2 t1 (3A)  β3 Bank reputation (or Lending relation) dummyt1  εt ,

Newline _ increase dummyt     Drawdown dummy   Covenant violation dummy 0 1 t1 2 t1 (3B)   3 Bank reputation (or Lending relation) dummyt1   4 Log(assets)t1

  5Leveraget   6ROAt  εt .

Equation (3A) includes a drawdown dummy, a covenant violation dummy, and bank reputation

(or prior lending relationship) indicator to explain whether a firm obtains a new line of credit within two years after the existing credit line expires. Equation (3B) includes a drawdown dummy, a covenant violation dummy, bank reputation (or prior lending relationship) indicator,

Log(assets), Leverage, and ROA to explain whether the rate on the new credit line is higher than that of the existing one. Drawdown is a dummy equal to one if a firm draws down its existing line credit before obtaining a new credit line, zero otherwise. Covenant violation is a dummy for whether there is a covenant violation between the year new line of credit was issued

(or two years after the existing credit line expires if new line of credit is not available) and the year the existing credit line was issued. Leverage (ROA) is the change in leverage ratio (ROA) between the year new line of credit was issued (or two years after the existing credit line expires if new line of credit is not available) and the year the existing credit line was issued. The high bank reputation dummy equals one if the lead bank's market share is above the sample median, zero otherwise. The strong lending relation dummy equals one if the lead bank-borrower prior lending relationship is above the sample median, zero otherwise.23

Estimates of the Heckman probit models are reported in Table 6. As shown for models

(1), (3), and (5), the drawdown dummy is significantly positively related to the probability of obtaining a new line of credit, suggesting that drawdown firms are more likely to obtain new credit lines after the existing one expires than firms that do not draw down. Drawdown firms

23 Log(assets), bank reputation, and prior lending relationship are based on the existing line of credit. 23 may have persistent liquidity needs, thereby a higher demand for credit lines. It also suggests that banks do not penalize drawdown firms by denying the renewal of lines of credit. This is sensible since banks do not want to lose business and clients. We also find that covenant violation reduces the likelihood of having a new credit line, which is consistent with the finding in Sufi

(2009) and Campello, Giambona, Graham, and Campbell (2010). While high bank reputation significantly enhances the probability of obtaining new credit lines, prior lending relationship appears not significantly related to the likelihood of obtaining new lines.

As shown in models (2), (4), and (6), conditional on obtaining a new line of credit, drawing down the existing line of credit is associated with a significantly higher probability of a rate increase on a new line. This suggests that drawdown firms are penalized with a higher rate.

In addition, firms that borrow from high reputation banks face a lower probability of rate increase, suggesting bank reputation mitigates the post-drawdown penalty, thereby providing a more efficient insurance contract. However, the prior lending relationship is not significantly related to post-drawdown rate change.

In summary, we document that banks penalize borrowers after they draw down lines of credit by charging a higher rate on new credit lines. However, drawdown firms are not punished by being denied new lines.

4.4.2. Post-drawdown Penalty: Price and Non-Price Contract Terms on Lines of Credit

Post-drawdown penalties could manifest in various dimensions, including price terms

(loan spread and annual fees) and non-price contract terms. In this section, we test hypothesis H2 further by examining post-drawdown penalties in various contract dimensions, and how bank reputation and prior lending relationship affect these penalties. For this purpose, we take all drawdown events and identify all lines of credit issued before and after each drawdown event. 24

We examine how loan rates and other non-price terms of the lines of credit change before versus after drawdown, controlling for borrower financial condition.

We only keep the first drawdown event for a firm if that firm draws down its line of credit more than once, which leaves us 2,065 credit facilities issued by 345 companies.24 Sample statistics are reported in Table 7. There are 705 credit facilities issued before drawdown and

1360 issued after drawdowns.25 The mean and median loan spread specified in these contracts is

155 and 125 basis points over LIBOR, respectively for each dollar drawn off a credit line (all-in drawn spread). The mean and median total (fees and interest) annual spread over LIBOR for each dollar available under a credit line (all-in undrawn spread) is 29 and 25 basis points respectively. Secured dummy is equal to one if a line of credit is secured by collateral, zero otherwise. The mean secured dummy indicates that about 67% of facilities include at least one collateral. Average time to maturity of these lines of credit is 43 months, with an average size of

$250 million. About 45% of facilities contain a performance pricing provision. To examine the effect of drawdown events on contract terms of lines of credit that are issued before and after the drawdown, we estimate the following regression model:

Loan contract term  α1  β1Post- drawdown  γiControl variablest-1  εt . (4) i

In equation (4), Post-drawdown is a dummy variable which equals one if a line of credit is initiated after the firm draws down its existing line of credit for the first time during our sample period. This key variable reflects the change in contract terms due to a drawdown event after we control for the change in firm and contract characteristics. Since penalty could take many forms,

24 We keep the first drawdown for each firm because we aim to examine the change of loan terms before and after drawdown events, and the post-drawdown window of the first drawdown is likely to overlap with the pre-drawdown window of second drawdown. This would be problematic in interpreting the results. 25 To be included in our sample, we require a firm to have at least one line of credit issued before and after the drawdown. 25 we will examine both loan price (all-in drawn spread and all-in undrawn spread) and non-price terms (time to maturity and collateral) of lines of credit before and after a drawdown.

Table 8 presents results of multiple regressions that examine the effect of drawdown events on line of credit price and non-price terms. Model (1) is an OLS regression with Log(All- in drawn spread) as dependent variable. The coefficient estimate on Post-drawdown is positive and statistically significant. Based on the magnitude of the coefficient estimate, the spread on lines of credit increases by 27 basis points after drawdown, holding all other variables at median values. It suggests borrowers must pay a significantly higher interest rate on lines of credit obtained after a drawdown. As shown in model (2), annual fees and interest charged on total available credit line (all-in undrawn spread) are significantly higher on lines of credit issued after a drawdown than those issued before the drawdown.26 Model (3) is an OLS regression to explain the logarithm of the time to maturity of lines of credit. We find that time to maturity decreases after the drawdown, though the result is not statistically significant. Model (4) is a Probit model with the secured dummy as dependent variable. The coefficient estimate on Post-drawdown is positive and significant at the 1% level, suggesting that the likelihood of including collateral in a new line of credit significantly increases after a firm draws down an existing facility. Based on the coefficient estimate of Post-drawdown, we find that, holding all other variables at their median values, borrowers are 15.73% more likely to pledge collateral on new lines of credit issued after a drawdown compared to facilities before the drawdown.

In general, our results show that drawing down a line of credit leads to significantly higher loan spreads, annual fees, and likelihood that collateral will be demanded on subsequent lines of credit. However, maturities are not significantly shortened on new credit lines. These results

26 Based on the magnitude of the coefficient estimate, undrawn spread of lines of credit increases by 3.69 basis points after drawdown compared to before the drawdown events, holding all other variables at their median values. 26 suggest that firms are punished by various contract dimensions after they draw down their lines of credit.

4.5. Bank Reputation, Lending Relationship, and the Penalty After the Drawdown

We document above that firms are punished after drawing down a line of credit. Next we examine whether bank reputation and prior lending relationship mitigate such penalties, as suggested by hypotheses H3 and H4. For this purpose, we include a High bank reputation dummy

(or Strong lending relation dummy) and its interaction with Post-drawdown in the regression models:

Loan contract terms  α1  β1High bank reputation  β2 Post - drawdown (5)  β3 High bank reputation*Post - drawdown  γiControl variablest-1  εt , i

Loan contract terms  α1  β1Strong lending relation  β2 Post - drawdown (6)  β3Strong lending relation*Post - drawdown  γiControl variablest-1  εt . i

In equation (5) or (6), 2 captures the effect of drawdown on contract terms for banks with low reputation or weak prior lending relationship, respectively. The impact of drawdown for banks with high reputation or strong prior lending relationship is captured by    in equation (5) or 2 3

(6) respectively, while 3 measures the differential effect of drawdown on contract terms between high versus low bank reputation, or strong versus weak lending relationship.

Table 9 reports the results of multiple regressions as shown in equation (5) and (6). In model (1), the coefficient estimate of Post-drawdown is positive and significant, suggesting that, for borrowers with low reputation lenders, the all-in drawn spread on newly-issued credit facilities rises significantly after the borrower uses an existing line of credit. However, the coefficient estimate of High bank reputation*Post-drawdown is negative and significant, leading

27 to an insignificant coefficient estimate for 2  3 . This indicates that firms borrowing from high reputation banks do not typically pay a higher loan spread on new credit facilities issued

after a drawdown. In model (2), the coefficient estimate of 3 is insignificant, suggesting no significant differential effect of drawdown associated with strong versus weak lending relationship. In models (3) and (4), we find that bank reputation but not prior lending relationship significantly reduces the positive association between drawdown and fees (all-in undrawn spread) on the lines of credit. While all-in undrawn spread is significantly increased in credit line contracts issued after a drawdown by low reputation bank, there is no significant change in all-in undrawn spread in contracts issued by high reputation banks. In models (5) and (6), we again find no significant effect of drawdown on the time to maturity of newly issued lines of credit, and no differential effect related to bank reputation and lending relationship. Model (7) shows that credit facilities issued after drawdown are more likely to include collateral, but only if the lender is low reputation. For borrowers with a high reputation lender, the impact of drawdown on the likelihood of including collateral becomes insignificant, as reflected in the coefficient estimate of . In contrast, the effect of drawdown on the probability of including collateral is no different in banks with strong versus weak prior lending relationship, as shown in model (8).

In summary, we find that while prior lending relation does not appear to mitigate penalties induced by drawdown, high reputation lenders are less likely to penalize borrowers, or penalize them to lesser extent, after they draw down an existing lines of credit. These results support our hypothesis H4: more reputable banks provide a more efficient liquidity insurance contract since reputation effect alleviates the moral hazard problem of lenders.

5. Conclusion

While the theoretical literature has long argued that lines of credit provide a liquidity insurance for firms to secure a desirable investment level in the event of future downturn

(Holmstrom and Tirole, 1998; Tirole, 2005), there is little empirical research on the extent to which a line of credit provides such insurance. In this paper we take advantage of a unique hand- collected dataset that allows us to directly measure the extent of insurance that a line of credit provides and examine how bank monitoring can mitigate moral hazard, thereby making a line of credit more efficient liquidity insurance.

Consistent with the liquidity insurance hypothesis, we find that the drawdown rate is significantly lower than the imputed market cost of borrowing given a firm's financial condition at the time of drawdown. Firms are more likely to draw down when they face a negative shock in financial performance. In addition, we document that borrowers are penalized (paying a higher loan spread and annual fee and more likely to pledge collateral) on new lines of credit issued after a drawdown. While stronger (or existence of) prior lending relation is associated with a lower drawdown rate (or larger drawdown rate benefit), borrowers are penalized much less by high reputation banks. It suggests that both bank reputation and lending relationship help provide more efficient liquidity insurance, though via distinct channel.

Our finding indicates that firms without strong bank relationship may not be able to benefit much from lines of credit. As a consequence, they should hold more cash instead.27 The liquidity that a line of credit can provide depends on future reward and punishment. This implies that in a serious recession when the survival of firms and banks is in question, there are greater limit to what a line of credit can provide. This is because the agency problems of both the banks

27 E.g., Sufi (2009), Lins, Servaes, and Tufano (2010). 29 and the firms become more severe as the future becomes more uncertain.28 More specifically, a firm on the verge of bankruptcy has the tendency to draw down the line of credit to stay in the business even if it knows that its chance of repay the loan is slim. On the other hand, the bank is more likely to decline a drawdown request or significantly increase the drawdown rate even if the borrower‟s project has a positive NPV. It is optimal for the bank to do so because maintaining a good reputation is less important than survival. Therefore our analysis suggests that the liquidity that lines of credit provide is pro-cyclical: it is low when the firms need it the most. This limit suggests the cash and line of credit may not be perfect substitutes for most firms.

28 Technically, it makes the parties less patient and thus more likely to behave opportunistically at the expense of long-term gain. 30

REFERENCES

Berger, Allen N., and Gregory F. Udell, 1995, Relationship lending and lines of credit in small firm finance, Journal of Business 68, 351-381.

Berkovitch, Elazar, and Stuart I. Greenbaum, 1991, The loan commitment as an optimal financing contract, Journal of Financial and Quantitative Analysis 26, 83-95.

Bharath, Sreedhar, Sandeep Dahiya, Anthony Saunders, and Anand Srinivasan, 2007, So what do I get? The bank's view of lending relationships, Journal of Financial Economics 85, 368-419.

Boot, Arnoud, Stuart I. Greenbaum, and Anjan V. Thakor, 1993, Reputation and discretion in financial contracting, American Economic Review 83, 1165-1183.

Boot, Arnoud, Anjan V. Thakor, and Gregory F. Udell, 1987, Competition, risk neutrality and loan commitments, Journal of Banking and Finance 11, 449-471.

Boot, Arnoud, Anjan V. Thakor, and Gregory F. Udell, 1991, Credible commitments, contract enforcement problems and banks: intermediation as credibility assurance, Journal of Banking and Finance 15, 605-632.

Campello, Murillo, Erasmo Giambona, John R. Graham, and Campbell R. Harvey, 2010, Liquidity management and corporate investment during a financial crisis, Working paper.

Carey, Mark, and Mark Hrycray, 1999, Credit flow, risk, and the role of private debt in capital structure, working paper.

Chemmanur, Thomas J., and Paolo Fulghieri, 1994, Reputation, renegotiation, and the choice between bank loans and publicly traded debt, Review of Financial Studies 7, 475-506.

Dinc, Serdar, 2000, Bank reputation, bank commitment, and the effects of competition in credit markets, Review of Financial Studies 13, 781-812.

Duca, John V., and David D. Vanhoose, 1990, Loan commitments and optimal monetary policy, Journal of Money, Credit and Banking 22, 178-194.

Fudenberg, Drew, and David K. Levine, 1992, Maintaining a reputation when strategies are imperfectly observed, Review of Economic Studies 59, 561-579.

Fudenberg, Drew, David I. Levine, and Eric Maskin, 1994, The folk theorem with imperfect public information, Econometrica 62, 997-1039.

Graham, John R., Si Li, and Jiaping Qiu, 2008, Corporate misreporting and bank loan contracting, Journal of Financial Economics 89, 44-61.

Heckman, James J., 1979, Sample selection bias as a specification error, Econometrica 47(1), 153-161. 31

Holmstrom, Bengt, and Jean Tirole, 1998, Private and public supply of liquidity, Journal of Political Economy106, 1-40.

Hoshi, Takeo, Anil Kashyap, and David Scharfstein, 1990, The role of banks in reducing the costs of financial distress in Japan, Journal of Financial Economics 27, 67-88.

Jimenez, Gabriel, Jose A. Lopez, and Jesus Saurina, 2009, Empirical analysis of corporate credit lines, Review of Financial Studies 22, 5069-5098.

Kaplan, Steven N., and Luigi Zingales, 1997, Do investment-cash flow sensitivities provide useful measures of financing constraints, Quarterly Journal of Economics 112, 169-215.

Kroszner, Randall. S., and Philip E. Strahan, 2001, Bankers on board: Monitoring, conflicts of interest, and lender liability, Journal of Financial Economics 62, 415-452.

Levin, Jonathan, 2003, Relational Incentive Contracts, American Economic Review 93, 835-857.

Lins, Karl V., Henri Servaes and Peter Tufano, 2010, What drives corporate liquidity? An international survey of cash holdings and lines of credit, Journal of Financial Economics 98, 160-176.

Myers, Stewart C., 1977, Determinants of corporate borrowing, Journal of Financial Economics 5, Issue 2, 147-175.

Nini, Greg, David C. Smith, and Amir Sufi, 2009, Creditor control rights and firm investment policy, Journal of Financial Economics 92, 400-420.

Petersen, Mitchell A., and Raghuram G. Rajan, 1994, The benefits of lending relationships: evidence from small business data, Journal of Finance 49, 3-37.

Roberts, Michael, and Amir Sufi, 2009, Renegotiation of financial contracts: evidence from private credit agreements, Journal of Financial Economics, forthcoming.

Rothschild, Michael, Joseph E. Stiglitz, 1976, Equilibrium in competitive insurance markets: An essay on the economics of imperfect information, Quarterly Journal of Economics 90, 630-649.

Sufi, Amir, 2007, Information asymmetry and financing arrangements: evidence from syndicated loans, Journal of Finance 62, 629-668.

Sufi, Amir, 2009, Bank lines of credit in corporate finance: an empirical analysis, Review of Financial Studies 22, 1057-1088.

Tirole, Jean, 2005, The theory of corporate finance, Princeton University Press.

Tirole, Jean, 2010, Illiquidity and all its friends, Journal of Economic Literature, forthcoming.

Appendix A: Variable Definitions

Variable Definition

Loan Characteristics Loan spread Spread over LIBOR rate in basis points Log(loan size) Natural logarithm of loan amount in US dollars Log(loan maturity) Natural logarithm of time to maturity of a loan in months Secured A dummy variable taking a value of 1 if the loan is secured, and 0 otherwise. Performance pricing A dummy variable taking a value of 1 if a performance pricing provision is included in a loan contract, and 0 otherwise. Upfront fee (basis points) A fee paid by the borrower upon closing of a loan Annual fee (basis points) Annual charge against the entire loan amount Loan purpose dummies Five dummies for various loan purposes, including corporate purposes, debt repayment, working capital, takeover, and all other purposes

Borrower Characteristics Log(assets) Natural logarithm of book value of assets Market-to-book The sum of market value of equity and book value of debt divided by book value of total assets Leverage Total debt divided by total assets ROA Net income divided by total assets Industry adj-ROA Change in ROA from the previous year minus the median change in ROA of the same two-digit SIC coded industry Tangibility Net PP&E divided by total assets Cash flow volatility Cash flow volatility that is computed as the standard deviation of net cash flow over the past 16 quarters divided by the average book assets over the same period Zscore Altman (1968) Z-score that is computed based on Compustat data items according to the following formula: 1.2*data179/data6 + 1.4*data36/data6 + 3.3*data178/data6 + 0.6*(data199*data25/data181) + data12/data6 Cash flow Net income divided by total assets Cash holding Cash and short-term investments divided by total assets Large firm A dummy variable taking a value of 1 if the firm‟s sales revenue are at least $1 billion, and 0 otherwise Investment grade A dummy variable taking a value of 1 if the firm has a rating of BBB− or higher, and 0 otherwise Stock return volatility The standard deviation of 12 monthly returns in previous year. No public debt dummy A dummy variable taking a value of 1 if a firm does not have any type of S&P debt rating available in the Compustat database, and 0 otherwise. 33

Lender Characteristics Lead bank market share (bank Dollar amount of loans arranged by the lead bank in a reputation) previous 5 years divided by the total amount of loans issued in the entire market in during the same period. Lead bank-borrower lending For a borrower j in a particular year, it is the amount of relation loans arranged by bank i and its predecessors during the previous 5 years divided by the total amount of loans borrowed by borrower j during the same period.

Macroeconomic factors Credit spread Yield difference between AAA rated corporate bond and BAA rated corporate bond Term spread Difference between 10-year Treasury yield and 2-year Treasury yield

Table 1: Sample statistics

Our sample consists of 800 randomly selected firms during 1996 to 2005. For each sample firm, we examine its 10-K filling at each fiscal year end and collect the following information: whether a firm has a line of credit, whether a firm draws down its line of credit, what the drawdown rate is, and whether a firm violates any financial or general covenants of its debt contracts.

Variable N Mean Median STD Having credit lines dummy 6859 0.831 1.000 0.375 Drawdown credit lines dummy 6777 0.526 1.000 0.499 Covenant violation dummy 6859 0.069 0.000 0.253 Asset ($ billion) 5656 1.535 0.277 3.754 Leverage 5704 0.243 0.234 0.196 Market-to-book 5539 1.862 1.442 1.275 ROA 5640 0.097 0.118 0.144 Tangibility 5656 0.296 0.214 0.239 Zscore 5318 4.213 3.085 4.758

Table 2: Univariate analysis on drawdown rate, predicted market rate, and contract rate on the line of credit

This table reports univariate results on drawdown rate, predicted market rate, and contract rate on the line of credit (LC) being drawn down and the new LC issued after the drawdown. All rates are expressed as basis points over LIBOR rate. T-tests and M-tests are used to test the null hypothesis that the mean and median difference is significantly different from zero respectively. P-values of these tests are reported in parentheses. ***, ** and * denotes significance at the 1%, 5% and 10% levels respectively.

Panel A: Summary statistics N Mean Median Min Max STD Contract rate on LC being drawn down 804 174.59 150.00 15.00 580.00 106.86 Drawdown rate 804 195.70 175.00 2.38 765.23 134.36 Predicted (imputed) market rate 631 223.10 221.66 4.35 479.48 87.79 Contract rate on LC after drawdown 606 177.02 150.00 27.50 580.00 106.87 Upfront fee on LC being drawn down 201 40.81 35.00 1.56 250.00 36.93 Annual fee on LC being drawn down 181 21.78 15.00 2.22 133.33 23.31 Upfront fee on LC after drawdown 95 42.91 25.00 5.00 200.00 47.30 Annual fee on LC after drawdown 133 21.60 17.50 6.25 133.33 18.67

Panel B: Difference between drawdown rate and predicted market rate N Mean Median (Drawdown rate – Predicted market rate) 631 -25.05*** -40.85*** (<.0001) (<.0001) (Drawdown rate – Contract rate on LC being drawn down) 804 21.11*** 0.00 (<.0001) (1.0000) (Drawdown rate – Contract rate on LC issued after drawdown) 606 3.80 2.00 (0.3771) (0.2164)

Table 3: Effect of bank reputation and prior lending relation on drawdown rate: univariate analysis

This table reports the univariate analysis of the difference between drawdown rate and predicted market rate (Drawdown rate – Predicted market rate) conditional on bank reputation and prior lending relation. Bank reputation is considered high if lead bank's market share is above the sample median, and considered lower otherwise. Bank-borrower prior lending relation is considered strong if the percentage of the amount of loans a firm borrowed from a particular bank during the previous five years is above the sample median, and considered weak otherwise. T-test, M-test, and F-test, and Z-test are used to test the null hypothesis that the mean, median, difference in mean, and difference in median is significantly difference from zero respectively. P-values of these tests are reported in parentheses. ***, ** and * denotes significance at the 1%, 5% and 10% levels respectively.

N Mean Median Bank reputation High 223 -37.97*** -54.39*** (<.0001) (<.0001) Low 231 -25.19*** -47.47*** (0.0085) (0.0024) Difference -12.78 -7.92 (0.3339) (0.5760) Bank-borrower prior Strong 225 -62.60*** -73.79*** lending relation (<.0001) (<.0001) Weak 229 -0.88 -11.53 (0.9226) (0.2904) Difference -61.72*** -62.26*** (<.0001) (<.0001) Existence of prior lending Yes 305 -50.35*** -62.66*** relation (<.0001) (<.0001) No 149 7.18 -4.11 (0.5326) (0.8699) Difference -57.53*** -58.55*** (<.0001) (<.0001)

Table 4: Effect of bank reputation and prior lending relation on the likelihood of drawdown and the drawdown rate: Heckman selection models

This table reports the results of Heckman selection models examining the effect of bank reputation and prior lending relation on the likelihood of drawdown and the drawdown rate. The dependent variable in the selection model is a binary variable that is equal to one if a borrower draws down its line of credit at a fiscal year end, and zero otherwise. Conditional on the drawdown, we examine how bank reputation and prior lending relation affect the difference between drawdown rate and the predicted market rate. Covenant violation is a dummy variable, which equals one if there is covenant violation in the drawdown year, and zero otherwise. High bank reputation dummy is equal to one if lead bank's market share is above the sample median, and zero otherwise. Strong lending relation dummy is equal to one if lead bank-borrower prior lending relation is above the sample median, and zero otherwise. All other variables are as defined in Appendix A. P-values are reported in parentheses below each coefficient estimate.

Panel A: Dependent variable = drawdown dummy Correct for endogenous choice of banks Variable (1) (2) (3) (4) (5) (6) High bank reputation dummy -0.009 -0.009 0.115 (0.917) (0.915) (0.174) Strong lending relation dummy -0.007 -0.008 -0.005 (0.928) (0.925) (0.955) Existence of prior lending relation -0.160 (0.086) Covenant violation 0.579 0.579 0.557 0.578 0.584 0.575 (0.002) (0.002) (0.003) (0.002) (0.002) (0.002) Industry adj-ROA -1.656 -1.655 -1.691 -1.656 -1.611 -1.652 (0.011) (0.011) (0.009) (0.011) (0.014) (0.011) Cash holding -5.059 -5.053 -5.025 -5.057 -5.060 -5.072 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Large firm -0.991 -0.993 -0.972 -0.991 -0.993 -0.972 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Investment grade -0.080 -0.083 -0.070 -0.081 -0.104 -0.090 (0.403) (0.381) (0.461) (0.401) (0.274) (0.340) Market-to-book 0.031 0.031 0.026 0.031 0.027 0.029 (0.522) (0.524) (0.589) (0.522) (0.569) (0.553)

Panel B: Dependent variable = drawdown rate - predicted market rate Correct for endogenous choice of banks Variable (1) (2) (3) (4) (5) (6) High bank reputation dummy 22.700 20.305 -5.234 (0.097) (0.131) (0.677) Strong lending relation dummy -50.932 -50.114 -55.742 (0.000) (0.000) (0.000) Existence of prior lending relation -51.584 (0.000) Covenant violation 61.996 44.923 47.246 47.842 58.924 44.644 (0.006) (0.043) (0.033) (0.032) (0.009) (0.044) First drawdown dummy 14.211 14.783 16.929 14.228 14.834 15.151 (0.252) (0.225) (0.166) (0.242) (0.233) (0.212) Log(assets) 5.379 8.467 7.985 5.736 8.450 8.744 (0.560) (0.341) (0.369) (0.527) (0.353) (0.323) Market-to-book 3.043 -2.875 -4.807 -1.548 1.532 -2.061 (0.750) (0.760) (0.613) (0.870) (0.873) (0.826) Leverage 146.919 146.694 162.476 144.127 150.450 144.732 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Industry adj-ROA 164.986 145.159 127.089 149.915 155.947 140.439 (0.068) (0.103) (0.156) (0.091) (0.086) (0.113) Tangibility 20.293 34.792 33.096 30.391 26.507 31.095 (0.509) (0.249) (0.273) (0.315) (0.389) (0.299) Zscore -3.978 -2.255 -2.071 -2.593 -3.670 -2.525 (0.240) (0.500) (0.537) (0.437) (0.279) (0.447) Log(loan maturity) -34.167 -38.464 -41.665 -37.964 -34.673 -38.941 (0.018) (0.007) (0.004) (0.008) (0.017) (0.006) Log(loan size) -33.075 -30.730 -29.778 -31.260 -32.156 -30.794 (0.000) (0.000) (0.001) (0.000) (0.000) (0.000) Performance pricing -10.720 -7.210 -6.403 -10.295 -6.729 -7.883 (0.501) (0.642) (0.681) (0.510) (0.672) (0.610) Loan purpose & industry dummy Yes Yes Yes Yes Yes Yes Inverse Mills Ratio -1.975 4.024 5.502 3.932 -.478 6.817 (0.921) (0.839) (0.784) (0.842) (0.981) (0.732) NOBS 1,151 1,151 1,151 1,151 1,139 1,139

Table 5: Effect of bank reputation and prior lending relation in the presence of high versus low information asymmetry

This table reports the results of Heckman selection models examining the effect of bank reputation and prior lending relation on the likelihood of drawdown and the drawdown rate, conditional on information asymmetry. The dependent variable in the selection model is a binary variable that is equal to one if a borrower draws down its line of credit at a fiscal year end, and zero otherwise. Conditional on the drawdown, we examine how bank reputation and prior lending relation affect the difference between drawdown rate and the predicted market rate. Information asymmetry is proxied by firms' access to public debt market and stock return volatility. No public debt dummy is equal to one if a firm does not have any type of S&P debt rating available in the Compustat database, and zero otherwise. High return volatility dummy is equal to one for firms with stock return volatility above the sample median of a particular year, and zero otherwise. Covenant violation is a dummy variable, which equals one if there is covenant violation in the drawdown year, and zero otherwise. High bank reputation dummy is equal to one if lead bank's market share is above the sample median, and zero otherwise. Strong lending relation dummy is equal to one if lead bank-borrower prior lending relation is above the sample median, and zero otherwise. All other variables are as defined in Appendix A. P-values are reported in parentheses below each coefficient estimate.

Panel A: Dependent variable = drawdown dummy Variable (1) (2) (3) (4) High bank reputation dummy 0.113 -0.014 (0.430) (0.907) Strong lending relation dummy -0.119 0.077 (0.380) (0.490) No public debt dummy 0.642 0.478 (0.000) (0.000) High return volatility dummy 0.027 0.162 (0.828) (0.177) High bank reputation×No public debt -0.153 (0.390) Strong lending relation×No public debt 0.158 (0.359) High bank reputation×High return volatility 0.057 (0.743) Strong lending relation×High return volatility -0.239 (0.166) Covenant violation 0.480 0.666 0.485 0.650 (0.011) (0.001) (0.010) (0.001) Industry adj-ROA -1.813 -1.877 -1.764 -1.892 (0.006) (0.006) (0.008) (0.006) Cash holding -5.533 -4.977 -5.530 -4.932 (0.000) (0.000) (0.000) (0.000) Large firm -0.805 -0.984 -0.796 -0.978 (0.000) (0.000) (0.000) (0.000) Investment grade 0.179 -0.071 0.172 -0.071 (0.100) (0.469) (0.111) (0.460) Market-to-book 0.022 0.024 0.022 0.021 (0.641) (0.632) (0.646) (0.669)

Panel B: Dependent variable = drawdown rate - predicted market rate Variable (1) (2) (3) (4) High bank reputation dummy 14.489 8.344 (0.539) (0.647) Strong lending relation dummy -17.541 -20.856 (0.430) (0.225) No public debt dummy -24.997 9.791 (0.341) (0.642) High return volatility dummy 42.808 79.108 (0.015) (0.000) High bank reputation×No public debt 12.356 (0.662) Strong lending relation×No public debt -47.896 (0.073) High bank reputation×High return volatility 28.658 (0.228) Strong lending relation×High return volatility -51.734 (0.031) Covenant violation 64.806 57.041 43.062 41.182 (0.003) (0.011) (0.050) (0.064) First drawdown dummy 12.676 13.010 13.294 11.998 (0.308) (0.295) (0.274) (0.326) Log(assets) 1.911 10.291 7.231 12.164 (0.840) (0.272) (0.423) (0.177) Market-to-book 2.697 0.071 -2.818 -4.868 (0.778) (0.994) (0.764) (0.607) Leverage 137.460 152.849 138.453 138.073 (0.001) (0.000) (0.001) (0.001) Industry adj-ROA 168.574 165.957 135.602 142.208 (0.064) (0.069) (0.128) (0.114) Tangibility 18.940 20.175 26.976 31.829 (0.538) (0.511) (0.374) (0.291) Zscore -4.053 -1.187 -2.125 -0.404 (0.232) (0.726) (0.523) (0.904) Log(loan maturity) -33.586 -24.338 -42.468 -31.585 (0.022) (0.104) (0.003) (0.033) Log(loan size) -33.100 -33.737 -31.685 -30.607 (0.000) (0.000) (0.000) (0.000) Performance pricing -10.260 -5.308 -3.230 -5.126 (0.519) (0.745) (0.836) (0.747) Loan purpose & industry dummy Yes Yes Yes Yes Inverse Mills Ratio 1.328 5.269 1.867 10.589 (0.950) (0.797) (0.928) (0.603) NOBS 1,151 1,092 1,151 1,092

Table 6: Effect of drawdown on the likelihood of obtaining a new line of credit and the change of contract rate: Heckman probit models

This table reports the results of Heckman probit models examining the effect of drawing down a line of credit on a firm's probability of obtaining a new credit line post-drawdown as well as the change of contract rate conditional on obtaining a new credit line. For this purpose, we start with all the lines of credit in the Dealscan database issued by our 800 randomly selected firms. For each line of credit, we try to identify whether the firm issued a new line of credit within two years after the expiration of the existing credit line. If the firm does, a dummy variable "newline" is set to one, and zero otherwise. A dummy variable "newline_increase" is equal to one if the contract rate on the new line of credit is higher than that of the existing one, and zero otherwise. The dependent variable in the selection model is a binary variable newline. Conditional on having a new line of credit, we examine how the drawdown event affects the likelihood of having a rate increase. Drawdown is a dummy variable, which equals one if a firm draws down its existing line credit before obtaining a new credit line, and zero otherwise. Covenant violation is a dummy variable, which equals one if there is a covenant violation between the year new line of credit was issued (or two years after the existing credit line expires when new line of credit is not available) and the year the existing credit line was issued, and zero otherwise. Leverage (ROA) is the change in leverage ratio (ROA) between the year new line of credit was issued (or two years after the existing credit line expires when new line of credit is not available) and the year the existing credit line was issued. High bank reputation dummy is equal to one if lead bank's market share is above the sample median, and zero otherwise. Strong lending relation dummy is equal to one if lead bank-borrower prior lending relation is above the sample median, and zero otherwise. Log(assets), bank reputation, and prior lending relation are based on the existing line of credit. All other variables are as defined in Appendix A. P-values are reported in parentheses below each coefficient estimate.

Newline Newline_ Newline Newline_ Newline Newline_ dummy increase dummy increase dummy increase dummy dummy dummy (1) (2) (3) (4) (5) (6) Drawdown dummy 0.250 0.420 0.271 0.395 0.259 0.445 (0.000) (0.007) (0.001) (0.024) (0.001) (0.010) Covenant violation -0.476 0.360 -0.430 0.432 -0.467 0.365 (0.000) (0.239) (0.000) (0.005) (0.000) (0.295) Log(assets) 0.062 0.084 0.072 (0.000) (0.000) (0.001) Leverage 0.636 0.508 0.503 (0.001) (0.027) (0.026) ROA -1.041 -1.421 -1.500 (0.004) (0.010) (0.004) High bank reputation dummy 0.297 -0.204 (0.000) (0.055) Strong lending relation dummy -0.009 0.104 (0.896) (0.104) NOBS 2,802 2,055 2,055

Table 7: Summary statistics of bank loans issued before and after the drawdown events

The sample consists of 2,065 Dealscan loans issued by 345 firms from 1990 to 2007 that have data available in Compustat and CRSP. To be included in this sample, we require a firm to have at least one line of credit before and after the drawdown event. All variables are as defined in Appendix A.

N Mean Median STD All-in drawn spread (basis points) 1853 155.06 125.00 110.82 All-in undrawn spread (basis points) 2060 29.29 25.00 18.44 Secured dummy 1247 0.67 1.00 0.47 Loan Maturity (months) 1894 42.49 41.85 22.78 Loan size ($ million) 2065 249.57 130.00 387.08 Performance pricing dummy 2065 0.45 0.00 0.50 Covenant violation dummy 1430 0.07 0.00 0.25 Log(assets) 2065 6.52 6.52 1.64 Market-to-book 2065 1.71 1.46 1.39 Leverage 2065 0.30 0.29 0.20 ROA 2065 0.13 0.13 0.09 Tangibility 2065 0.35 0.28 0.25 Cash flow volatility 2065 0.06 0.05 0.03 Zscore 2065 3.47 2.85 5.45

Table 8: Effect of drawdown events on price and non-price contract terms of lines of credit

This table reports regression results explaining the effect of drawdown on various contract terms of credit facilities. Models (1) - (3) are OLS regressions with Log(All-in drawn spread), Log(All- in undrawn spread), and Log(maturity) as the dependent variables, respectively. Model (4) is a Probit regression explaining the probability of a facility being secured. Post-drawdown is a dummy variable, which equals one if a facility is issued after the first drawdown year, and zero otherwise. Covenant violation is a dummy variable, which equals one if a firm violates its debt covenant in a particular year, and zero otherwise. All other variables are as defined in Appendix A. P-values are reported in parentheses below each coefficient estimate.

Log(All-in Log(All-in Log(maturity) Secured dummy Variable drawn spread) undrawn spread) (1) (2) (3) (4) Post-drawdown 0.190 0.136 -0.065 0.443 (0.000) (0.001) (0.151) (0.001) Covenant violation 0.235 0.109 -0.099 0.744 (0.000) (0.117) (0.159) (0.004) Log(assets) -0.168 -0.123 -0.144 -0.188 (0.000) (0.000) (0.000) (0.003) Market-to-book -0.123 -0.110 -0.066 -0.002 (0.000) (0.000) (0.019) (0.984) Leverage 1.076 0.933 0.276 1.999 (0.000) (0.000) (0.005) (0.000) ROA -1.092 -0.831 0.184 -1.188 (0.000) (0.000) (0.321) (0.071) Tangibility -0.190 -0.277 0.059 -0.312 (0.016) (0.001) (0.496) (0.220) Cash flow volatility 0.654 -0.177 -1.134 2.166 (0.204) (0.739) (0.041) (0.220) Zscore 0.027 0.019 0.009 0.004 (0.000) (0.007) (0.206) (0.854) Log(loan maturity) 0.086 0.189 0.325 (0.001) (0.000) (0.000) Log(loan size) -0.121 -0.090 0.215 -0.294 (0.000) (0.000) (0.000) (0.000) Performance pricing -0.020 0.071 0.231 -0.360 (0.581) (0.055) (0.000) (0.005) Term Spread 0.101 0.075 -0.079 0.060 (0.000) (0.001) (0.002) (0.397) Credit spread -0.192 -0.063 0.091 0.014 (0.029) (0.475) (0.345) (0.959) Loan purpose & industry Yes Yes Yes Yes dummy NOBS 1258 1,100 1333 899 Adj R2 or Pseudo R2 0.5079 0.438 0.2332 0.2714

Table 9: Effect of drawdown events on the price and non-price contract terms of lines of credit: the moderate effect of bank reputation and prior lending relation

This table reports regression results examining how bank reputation and prior lending relation moderate the effect of the drawdown events on various contract terms of credit facilities. Models (1) - (6) are OLS regressions with Log(All-in drawn spread), Log(All-in undrawn spread), and Log(maturity) as the dependent variables,, respectively. Models (7)-(8) are Probit regressions explaining the probability of a facility being secured. Post-drawdown is a dummy variable, which equals one if the line of credit is initiated after the first drawdown, and zero otherwise. High bank reputation dummy is equal to one if lead bank's market share is above the sample median, and zero otherwise. Strong lending relation dummy is equal to one if lead bank-borrower prior lending relation is above the sample median, and zero otherwise. Covenant violation is a dummy variable, which equals one if a firm violates its debt covenant in a particular year, and zero otherwise. All other variables are as defined in Appendix A. P-values are reported in parentheses below each coefficient estimate.

Log(All-in drawn Log(All-in undrawn Log(maturity) Secured dummy spread) spread) Variable (1) (2) (3) (4) (5) (6) (7) (8)

Post-drawdown 2  0.342 0.116 0.287 0.050 -0.103 -0.044 0.730 0.369 (0.000) (0.077) (0.000) (0.441) (0.151) (0.542) (0.000) (0.069)

High bank reputation 1  0.377 0.328 -0.027 0.910 (0.000) (0.000) (0.768) (0.000) High bank reputation × Post- -0.376 -0.338 0.042 -0.897 drawdown 2  3  (0.000) (0.000) (0.677) (0.001) Strong lending relation -0.185 -0.200 0.086 -0.054 (0.026) (0.014) (0.337) (0.824) Strong lending relation ×Post- 0.089 0.143 -0.077 -0.113 drawdown (0.331) (0.111) (0.435) (0.678) Covenant violation 0.234 0.250 0.103 0.122 -0.050 -0.050 0.675 0.803 (0.001) (0.001) (0.179) (0.111) (0.532) (0.533) (0.019) (0.006) Log(assets) -0.184 -0.185 -0.151 -0.151 -0.164 -0.163 -0.243 -0.235 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.001) (0.001) Market-to-book -0.129 -0.121 -0.134 -0.124 -0.051 -0.052 -0.082 -0.051 (0.000) (0.000) (0.000) (0.000) (0.104) (0.102) (0.382) (0.574) Leverage 0.855 0.875 0.859 0.871 0.222 0.219 2.106 2.110 (0.000) (0.000) (0.000) (0.000) (0.060) (0.064) (0.000) (0.000) ROA -1.231 -1.235 -0.815 -0.857 0.221 0.227 -0.611 -0.687 (0.000) (0.000) (0.002) (0.001) (0.370) (0.357) (0.431) (0.374) Tangibility -0.190 -0.183 -0.227 -0.216 0.003 0.002 -0.266 -0.190 (0.027) (0.033) (0.009) (0.013) (0.977) (0.982) (0.347) (0.498) Cash flow volatility 1.119 0.924 -0.015 -0.185 -1.279 -1.265 1.968 1.152 (0.063) (0.127) (0.979) (0.756) (0.049) (0.051) (0.316) (0.544) Zscore -0.009 -0.005 0.005 0.007 -0.003 -0.003 -0.015 -0.010 (0.419) (0.612) (0.628) (0.509) (0.825) (0.780) (0.672) (0.759) Log(loan maturity) 0.098 0.099 0.187 0.190 0.395 0.386 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Log(loan size) -0.106 -0.099 -0.073 -0.068 0.225 0.226 -0.297 -0.256 (0.000) (0.000) (0.004) (0.008) (0.000) (0.000) (0.000) (0.002) Performance pricing -0.018 -0.012 0.059 0.067 0.236 0.236 -0.450 -0.457 (0.632) (0.750) (0.145) (0.099) (0.000) (0.000) (0.002) (0.002) Term Spread 0.098 0.102 0.080 0.084 -0.078 -0.078 0.088 0.095 (0.000) (0.000) (0.001) (0.001) (0.004) (0.004) (0.261) (0.222) Credit spread -0.254 -0.245 -0.105 -0.097 0.163 0.163 -0.081 -0.099 (0.007) (0.009) (0.264) (0.307) (0.119) (0.119) (0.791) (0.743) Loan purpose & industry dummy Yes Yes Yes Yes Yes Yes Yes Yes

2  3 -0.034 0.205 -0.051 0.193 -0.061 -0.121 -0.167 0.256

P-value (H0: 2  3  0 ) (0.596) (0.002) (0.431) (0.003) (0.397) (0.089) (0.417) (0.187) NOBS 1073 1073 946 946 1128 1128 753 753 Adj R2 or Pseudo R2 0.509 0.505 0.458 0.453 0.244 0.244 0.284 0.273